Metadata: Everyone Talks About It, But What Is It?

Data Warehousing and Solutions
Paper 125-26
Metadata: Everyone Talks About It, But What Is It?
John E. Bentley, First Union National Bank
Abstract:
Everyone agrees that metadata is important. But then why
do so many data warehouse and data mart users complain
about the unavailability or completeness of their metadata?
At least part of the reason is that there’s no agreement on
just exactly what is “metadata”. The simple definition—“data
about data”—is too fuzzy and in most contexts simply
inadequate. Unfortunately, most contextual-specific
definitions are unworkable or inappropriate for business
users. This paper provides a definition that can be
incorporated into a metadata solution for those who often
need it most—the business users.
Disclaimer: The views and opinions expressed here are
those of the author and not those of First Union National
Bank. First Union National Bank does not necessarily
subscribe to any philosophy, school of thought, approach,
definition or process the author describes.
Why is Metadata a Problem?
Although metadata is widely acknowledged as critical for
getting the most business value out of a data warehouse (in
this paper the term “data warehouse” includes data marts),
most data warehouse managers actually do little than talk
about metadata. In a 1998 survey of 154 data warehouse
managers by The Data Warehousing Institute, only 25
percent of respondents had deployed or were deploying a
metadata solution. Twenty-one percent reported having
developed a plan but had not yet implemented it, and 54
percent of respondents reported that they had “no plans” to
implement a metadata solution.
Although these statistics are over two years old, there is no
hard evidence to show that the situation has improved
dramatically. Most anecdotal evidence suggests that little
has changed. It still appears likely that less than half the
companies that have a data warehouse are willing to invest
the time, money, and resources needed to implement a
comprehensive metadata management system, leaving
users with only partial, ad hoc solutions.
Part of the reason for this is false assumption that the return
on investment (ROI) for a metadata management system is
difficult to quantify in terms of increasing revenues or
decreasing expenses, but the costs in terms of people,
software and time are clear. (See Marco, 2000.) Data
warehouse projects are usually on an “aggressive” project
schedule, but collecting metadata is time consuming when
documentation for the legacy systems that feed the
warehouse is unorganized, sketchy, or simply unavailable.
Another reason is that the mindset and priorities of both the
technical staff and the business sponsor are different. Most
frequently, the business side has authority over the project—
they control the budget—and so the technical group has to
implement the project as directed. Because of up-front cost
or time constraints, the business sponsor may minimize the
metadata effort even over the advice of the technical staff.
As a result, metadata is assigned a “do it later” priority, but
later never quite comes.
At the same time that the business sponsors may be
neglecting the business metadata, an aggressive project
plan forces the technical staff to concentrate on the system
design, build, and initial load phases and not on
maintenance and use. The IT staff then quite naturally
focuses on the metadata that they need to do their job.
Because of these “organizational pathologies” and for other
reasons, most companies poorly document the source and
nature of the data they are warehousing. Even fewer
produce metadata that can be leveraged to automate or
improve extract-transform-load processes, data quality, or
production reporting tasks.
Why is Metadata Important?
An underlying, often-ignored truth is that a data warehouse
is only as good as its metadata. A data warehouse does not
generate real value until it is exploited to provide information
that supports business decision-making. Metadata is critical
to exploitation because it tells users (and programmers!)
where to find the exact data they need and helps them
understand what it means. Put simply, good metadata
makes it easier to use the data warehouse by allowing faster
turn-around for information requests. Faster turn-around
equals higher productivity, and ease-of-use gives users
confidence in the information retrieved.
Business users must have confidence in the data in the
warehouse and the answers it provides. Otherwise, they will
be disinclined to use any data except that with which they
are already confident—they will revert to using their own
islands of parochial data where they know and trust its
lineage. If that happens, the business value of the data
warehouse is lost. The next diagram shows the impact
metadata has on the decision-making process.
Diagram 1: Metadata in the Decision-Making Process
Requires
Data
Warehouse
Data Collection and
Transformation Procedures
Capture
Drives
Provides
Enables
Meta Data
Data Integration
Delivers
Ensures
Data Quality and Reliability
Consistent, Timely, and
a Full-Range of Data
Bolsters
Advanced Data Analysis
Techniques
One Stop
Information
Portal
Stimulates
Increases
Accelerates
Reinforces
Trust in the Data
Empowers
Informed Decision
Making
Data Accessibility
and Security
Adapted from Fletcher and Pinner, “Navigating the Data
Warehouse Paradox Zone.”
Data Warehousing and Solutions
Metadata provides real value to a data warehouse and
increases the ROI. David Marco, an industry-recognized
expert on metadata, has developed formulas for calculating
metadata ROI and estimates that good metadata can add
1% to a corporation’s revenues. (Marco, 2000.) Metadata
contributes measurable value by:
that are hard for non-technical managers to decipher. As
Table 1 shows, “data about data” has evolved into a number
of things, depending on its IT context.
Table 1: Metadata in Context
Context
Description
Data Administration
Properly documented logical
and physical data models and
entity-relationship diagrams for
both source and target systems.
Usually controlled by the IT staff
and usually a high priority.
Data Warehouse
Back-End
Documentation tracking the
extract-clean-transform-load
process. Source system,
staging area, and data
warehouse data structures,
copybook names, column
mapping translation tables, and
other ETL documentation.
Controlled by IT staff and
receives a high priority.
Application
Development
Documentation about
processes that access data via
an application. Presented as
process models or
decomposition diagrams. Also
includes pseudo-code and the
final program code with internal
documentation. Due to time
and resource constraints,
typically is ignored or given low
priority.
Our project is to provide contact information for Inactive
customers who are most likely to move into the Occasional
category. But notice that there’s no listing for category D—
what are we to make of that? Is D a valid category that
hasn’t yet made it into the metadata? Is it an old category of
“Dormant” that is no longer valid and should have been
recoded to “Inactive”? Does D mean “Dropped, had an online account but never used it”? Is it a legacy system default
value?
Data Warehouse
Front-End
Documents the meaning of data
for the benefit of those running
queries against the warehouse
and interpreting the results.
Imperative to ensuring an
accurate, consistent
interpretation of the data, but
often is assigned low priority.
The point is, we don’t know and until we get an answer these
161,000 records—about 6 percent of the number of Inactive
customers--are useless in our project. Do we ignore them
and possibly not contact customers we should be contacting,
include them in the selection at the risk of contacting
inappropriate customers, or spend time figuring out who
these customers are and why the data is like this? In effect,
we have 161,000 possibly relevant customer records in the
data warehouse that we can’t immediately use because the
metadata is incomplete. (This also points to a failure of data
quality assurance because metadata isn’t incorporated into
the QA process, but that’s a subject for another paper.)
In the list of contexts presented above, the data warehouse
front-end context is clearly the one most referenced by and
important to business users because it helps understand the
data. In an article about optimizing data warehouse usage,
one data warehouse guru used an analogy about pioneers
exploring the Wild West (Devlin 1988). Like the Wild West, a
data warehouse is a vast territory. Without maps, the
Western pioneers were often. Likewise, data warehouse
users will be lost without the map that metadata provides.
•
•
•
•
•
Improving decision making accuracy;
Reducing new employee training costs;
Increasing user confidence in the data warehouse,
which results in higher usage and productivity;
Allowing sophisticated data quality processing; and
Identifying mistakes and problems with source IT
systems
An Example of Why Metadata is Important for
Business Users
Consider the following scenario. Among other things, a data
set extracted from a warehouse contains a variable with a
business name of “Customer On-Line Access Category.”
The distribution is
A DI OP X -
540,000
161,000
2,690,000
1,859,000
126,000
18,930,000
The metadata shows that
A = Active, 3 month on-line average > 6 times
I = Inactive, not on-line in 3 months
O = Occasional, 3 month on-line average between
1 and 6 times
P = Pending, on-line account applied for but not
assigned
X = Not an on-line customer
Metadata Contexts and Views
There is a third reason for the disinclination to properly
address the metadata issue, and that reason has its roots in
how metadata is defined. The term “metadata” doesn’t has
a simple, universal meaning. “Data about data” no longer
works and different functional groups within IT have applied
their own definitions. As a result, “metadata” has become a
technical code word that, depending on the context in which
it is being used, generates conflicting definitions containing a
variety of sometimes vague and often misleading messages
The same data warehouse expert categorizes metadata as
build-time, usage, and control, and suggests two “views” of
metadata: builder and end-user.
Table 2. Metadata Views
View
Description
Builder
A blueprint. Key component is the enterprise
data model. The ultimate definition of what the
warehouse is.
End-User
A flexible, easy to use “route map”. Defining
feature is that the warehouse contents are
presented in a business context.
Data Warehousing and Solutions
So, What is Metadata for Business Users?
Some definitions of metadata appropriate for business users
do exist, but they range from broad to narrow in their scope.
•
•
•
•
“Metadata is high-level data that describes low-level
data. ... [It] maps the data to business concepts that
are familiar and useful to end-users.” (Korzybski,
March 1996.)
“In the context of data warehousing, the term refers to
anything that defines a data warehouse object, such as
a table, query, report, business rule, or transformation
algorithm. … It also supplies a blueprint that shows
how one type of information is derived from another.”
(Gardner, November 1997.)
“Metadata describes the information in the data
warehouse: what is means, where it came from, how it
was calculated, when it was loaded, who owns it.”
(Ekerson, March 2000.)
“Metadata is the definitions, sources, rules, and
thresholds used to constrain the business data you are
collecting, validating, transforming, reconciling, loading,
and reporting.” (Fletcher and Pinner, March 2000.)
table shows some specific examples of metadata grouped
into Kimball’s categories.
Table 3: Examples of Metadata
Category
BackEnd
Example
9
9
9
9
9
9
9
9
Metadata for Business Users
9
Technical metadata is used by IT professionals in the
planning, design, creation, and maintenance of the data
warehouse. This is the Data Administration, DW Back-End,
and Application Development contexts. But as is pointed out
in a SUGI25 paper, “[b]usiness users require more
descriptive information, which will assist in translating
codified information into the business concepts relevant to
their domain. This would include the content and purpose of
the data, related business rules, ownership and
administration, and location.” (Stevens, 2000)
9
9
9
9
9
With this in mind, here is a clear, concise definition of
“Business Metadata” based on the Data Warehouse FrontEnd context description:
Business metadata shows non-technical
users where to find information in the data
warehouse, where it came from and how it
got there, describes its quality, and provides
assistance on how to interpret it.
Technical (back-end) metadata and business (front-end)
metadata are mutually reinforcing. This definition is
purposely broad enough to include technical metadata
because many users want or need a deeper understanding
of the origin and evolution of the data.
Implicit in this definition is the assumption that metadata will
be dynamic, thereby helping ensure overall data quality and
reliability which will, in turn, bolster users trust in the data.
Dynamic metadata provides the ability to review the lineage
of the data. Decision-makers will know where the data came
from, how it was transformed, and what it really means.
What does Metadata Contain?
A complete metadata solution requires a lot of information.
Ralph Kimball bases his categorization of metadata on the
intended user base and distinguishes between “back-room
metadata” that guides the extraction, cleaning, and loading
processes and “front-room metadata” needed by query tools
and report writing. As mentioned earlier, although there is a
lot of crossover between back-end and front-end metadata,
it’s the front-room metadata that makes the data in the
warehouse really meaningful to business users. The next
FrontEnd
9
9
9
9
9
9
9
9
9
9
9
9
9
Ownership descriptions of each
source schemas
Source file layouts and target schema
designs
Definitions and characteristics of
tables and columns
Primary/foreign key assignment
scheme and relationships
Database partition and disk striping
specifications
Index and view definitions and
specifications
Mainframe or source system job
specifications
File/copy book descriptions and
specifications
COBOL/JCL, C or Basic code to
implement extraction
Update frequencies of the original
sources
Job specifications for joining sources,
stripping out fields, and looking up
attributes.
Data cleaning, enhancement, and
transformation rules, specifications,
and mappings
Data audit records and transformation
run-time logs
Access methods, access rights,
privileges and passwords for source
access
Process flows, e.g., BPwin
Presentation graphics, e.g.,
PowerPoint
Flowcharts and program code for
accessing source system data
Ownership and Business descriptions
of the source systems
Ownership and Business name of
data elements
Business-rule based definitions of
data elements
Descriptions of the valid values in
categorical fields
Descriptions and flowcharts of
aggregation and transformation
processes
Physical data models
Table join guidance, including
cautions and restrictions
Validation statistics for quality control
Legal limitations on usage
User login profiles and
security/access controls
Data Warehousing and Solutions
With these examples in mind and considering the different
contexts and views of metadata and the back-end/front-end
distinction, here is a list of some metadata items that are
essential for business users:
Table 4. Critical Business Metadata
All variables
¾ Variable name
¾
Variable business name
¾
Variable definition (short)
¾
Variable description (long)
¾
Data set name
¾
Data set business name
¾
Data set description
¾
Legacy system and Quality Assurance contacts
¾
Update frequency
¾
Date of last update
¾
Special missing values
¾
List of variable names used if the variable is a
created or calculated variable
¾
Business logic, algorithms, and pseudo-code used
in cleaning, transforming, creating, summarizing, or
calculating the variable
¾
Special cautions, legal limits, tips and clues on
usage
Categorical variables
¾ List of valid values and their definitions
¾
Frequency distribution including number of missing
values
Interval variables
¾ Formula used in calculating the variable
¾
Descriptive statistics including the number of
records, mean, standard deviation, median, number
of missing, and range.
Creating and maintaining metadata will require both up-front
and on-going investments of time and resources. For the
most part, though, the metadata will be static or slowly
changing and it’s creation can be automated, so the
resources needed for maintenance will comparatively small.
Calculating or generating the specific items needed for
categorical and interval variables should be included as part
of the warehouse load process. For this, when a very large
number of records are loaded a weighted sample may
suffice, depending on the accuracy needs of the users.
Metadata updating processes can easily be incorporated into
an automated or metadata-driven quality assurance process.
For example, PROC COMPARE makes it easy to compare
the current month’s list of valid values for a categorical
variable with last month’s list as a quick check for new
categories. Likewise, comparing the current month’s
summary descriptive statistics such as mean, standard
distribution and percentiles to last month’s will quickly
identify anomalies in the data.
Competing Metadata Standards and the SAS
Metadata Architecture
Until September 2000, companies working to implement
strong metadata management processes and procedures
were handicapped by competing metadata standards within
the IT industry. There are two competing groups, both
spearheaded by industry heavyweights. These groups are
The Meta Data Coalition (MDC) and The Object
Management Group (OMG). Competition going forward,
however, should be minimized because in September 2000
the two groups agreed to merge, retaining the OMG name,
and work together to develop a single best-of-breed
standard built around the OMG’s Common Warehouse
Metamodel (CWM).
The SAS Institute is a member of the 53-member Metadata
Coalition, as is Microsoft. The MDC sponsors the Open
Information Model (OIM) as a comprehensive source of
standards and specifications for business engineering,
knowledge management, and databases in addition to data
warehousing. The OIM schema consists of standard object
types and relationships described in Unified Modeling
Language. The OIM is vendor-neutral and uses SQL as a
query language and Extensible Markup Language (XML) as
an interchange format between data repositories. In October
2000 the MDC endorsed the Microsoft Repository as it’s
preferred metadata storage database.
Major members of the Object Management Group are IBM
and Oracle, among others. The OMG provids a standard
called the Common Warehouse Metamodel (CWM) to
enhance metadata sharing and interoperability in data
warehousing environments. The CWM is based on the
OMG’s UML (Unified Modeling Language), XML (XML
Metadata Interchange) and MOF (Meta Object Facility) and
incorporates the Common Object Request Broker
Architecture (CORBA) as the basis for interoperability and
application integration.
The merger of MDC into the OMG marks a major agreement
of data warehousing and metadata vendors to converge on
one standard, incorporating the best of the MDC's Open
Information Model with the OMG’s Common Warehouse
Metamodel. When the work is complete, the resulting
specification will be issued by the OMG as the next version
of the CWM. A single standard will allow users to freely
exchange metadata between different products from
different vendors.
SAS Software uses a layered metadata architectural
approach that maximizes flexibility. Four distinct functional
layers combine to minimize the need to continually develop
and revise the metadata architecture to account for changes
in metadata sources, application needs, and hardware
platforms.
Table 5. SAS Metadata Architecture
Layer
Description
Facility
The Common Metadata Facility (CMF)
provides tactical control. It controls the
creation, deletion, update, and
persistence of metadata objects.
Model
The Common Metadata Model (CMM)
ensures that applications that access the
metadata all interpret it the same way. It
defines the objects and relationships.
API
The Application Program Interface (API)
is the presentation layer and, in Version
8, may be written in either C or SCL.
Repository
The Repository stores the metadata.
Users can choose from many different
physical formats, and it can reside on
almost any platform.
Adopted from Vernee Stevens, “SAS Metadata Architecture
and Current Industry Metadata Trends.”
Data Warehousing and Solutions
Summary
The value of a data warehouse doesn’t come from having a
lot of data in one place. A data warehouse by itself has only
potential value. The return on investment is realized when
the data is converted to information and then used to make
decisions that solve problems. Without comprehensive
metadata, though, much of the data in a warehouse may
never become information; at best, converting it to
information will take much longer and even then decisionmakers may have doubts about its accuracy.
Unfortunately, metadata can be costly to compile and make
available to users. Metadata experts, however, have
developed guidelines for calculating metadata ROI, showing
that business and technical metadata can both increase
revenues and reduce expenses. For business users,
metadata improves the accessibility, quality, credibility, and
usability of a data warehouse in multiple ways:
•
•
•
•
•
By documenting flows of data used to populate the
warehouse;
By documenting the creation, calculation, and
summarization processes;
By providing dynamic quality-control metrics;
By allowing the data warehouse to be navigated using
meaningful business terms; and
By allowing the use of advanced data analysis
techniques needed for data mining.
The need for a complete metadata solution becomes
increasingly apparent as organizations begin to deploy
second and third generation decision support databases that
propagate data from the data warehouse into specialized
data marts. As the lineage of data increases and the
numbers of users grow, the need (and demand) metadata
multiplies.
The metadata needs of business users are different from the
technical staff, but there are numerous overlaps. A common
set of shared metadata that includes the critical items in
Table 4 will benefit both groups. For business users though,
the basic metadata they need is that which shows them
where to find information, where it came from and how it got
there, describes its quality, and provides assistance on how
to interpret it.
References and Resources
Devlin, Barry (1998) “Metadata: The Warehouse Atlas.” DB2
Magazine Online, Spring. www.db2mag.com/98spWare.htm
Ekerson, Wayne W. (2000) “Ignore Meta Data Strategy at
Your Peril.” Application Development Trends, March.
Fletcher, Tom and Jeff Pinner (2000) “Navigating the Data
Warehousing Paradoz Zone.” dmDirect, March 3.
www.dmreview.com
The Hurwitz Group (1988) Enterprise Metadata
Management, December. www.hurwitz.com
Kimball, Ralph (1998) “Meta Meta Data Data.” DBMS
Magazine, March. www.dbmsmag.com/9803d05.html
Korzybski, Alfred (1996). “What is Metadata?” Data
Warehousing Tools Bulletin, March 1.
www.computerwire.com/bulletinsuk/212e_1a6.htm
Marco, David (2000) Meta Data ROI 3-part series. DM
Review, September, October, November.
Marco, David (2000) Building and Managing the Meta Data
Repository: A Full Life-Cycle Guide. New York: Wiley.
Meyers, Rachel (1998) Metadata series. The Data
Warehousing Career Newsletter, July.
www.softwarejobs.com/dataware/7-10-98.html
www.softwarejobs.com/dataware/7-17-98.html
www.softwarejobs.com/dataware/7-31-98.html
Stevens, Vernee (2000). “SAS Metadata Architecture and
Current Industry Metadata Trends.” SUGI25 Proceedings.
Wiener, Jerry (2000) “Meta Data in Context.” dmDirect,
February 11. www.dmreview.com
Contact Information
John E. Bentley
First Union National Bank
201 S. College Street
Mailcode NC-1025
Charlotte NC 28288
704-383-2686
John.Bentley2@FirstUnion.Com
About the Author
John Bentley has used SAS Software for fourteen years in
the healthcare, insurance, and banking industries. For the
past three years he has been with the Enterprise Information
Group of First Union National Bank with responsibilities of
supporting users of First Union’s data warehouse and data
marts and managing the development of SAS client-server
applications to extract, manipulate, and present information
from it. John is regularly presents at national, regional, and
local SAS User Group Conferences and is on the Executive
Committee of the Data Mining SAS User Group.